Binary Classification of Faba Bean (Vicia faba L.) Cultivars Based on Appearances Using Image Processing Technique and Machine Learning Algorithms

Appearance is one of the important traits in seeds. Appearance-related features such as shape, size, and color are important parameters in distinguishing seeds from each other. Machine learning algorithms are used to distinguishing plant seed species for different purposes. In this study, four faba bean cultivars (Alexia, Alice, Jasmin, and Arabella) were used to distinguishing based on appearance measurements including shape and size features analyzed in pairs. Eleven machine learning algorithms (NB, MLP, SGD, SL, LMT, SMO, kNN, J48, Random Forest, Random Tree, REPTree) were used to assess binary classification performance utilizing red-green-blue (RGB) color channels through a image processing system. Among all pairs, faba bean seeds of the Alexia and Alice cultivars had the greatest classification accuracy of 90.5% using the Random Forest, and 87.5% with the MLP, SGD, and J48 models. The MLP model achieved the highest accuracy rate of 87% for the categorization of Alexia vs Arabella cultivars, followed by the J48 model with an accuracy rate of 84%. The Alice cultivar possesses the greatest values for area (83.80 mm²), perimeter (47.43 mm), width (9.28 mm), and length (12.50 mm). Wilks' lambda results indicated that the variations in external appearance of faba bean varieties are significant (p < 0.01). All of these results indicated that machine learning algorithms can effectively differentiate faba bean seeds based on their physical characteristics.